RSNA 2003 Scientific Papers > Computer-aided Diagnosis (CAD) in Rheumatoid Arthritis: ...
  Scientific Papers
  SESSION: Health Services, Policy and Research (Issues in Research Methodology)

Computer-aided Diagnosis (CAD) in Rheumatoid Arthritis: Automated Joint Localization, Estimation of the Bone Contour and Consecutive Detection of Defects of the Bone Contour in Metacarpal Bones

  DATE: Wednesday, December 03 2003
  START TIME: 11:40 AM
  END TIME: 11:47 AM
  CODE: K16-1003

Philipp Peloschek MD
Vienna Austria
Georg Langs
Horst Bischof
Franz Kainberger MD
Walter Kropatsch
Herwig Imhof MD

Arthritis, rheumatoid
Computers, diagnostic aid
Images, processing

Purpose: The aim of this project is the development of methods to perform an automated analysis of serial hand radiographs. Up to now our system allows automated detection of joints and delineation of bone contours resulting in a parametrical description of the identified shape. Automated discrimination of pathological changes of shape will be final step towards computer aided diagnosis (CAD) and quantification of erosions in rheumatoid arthritis.

Methods and Materials: The calculations are performed on a commercially available laptop on digital radiographs (DICOM). No manual initialisation is required for the localisation procedure. The training of the contour delineation algorithm was performed on two training sets T15 and T30 consisting of n=15 and n=30 sample radiographs. Statistical analysis for the localizer procedure was done on the CMC-, MCP- and DIP-joints of 10 subjects. Evaluation of the automated bone contour delineation was done using 42 landmarks per metacarpal bone on 10 different radiographs leading to a sample data set of 420 measurements. Standard of references was the bone contour definition of an experienced radiologist.

Results: Automated joint localiziation was found to be sufficiently robust (median position error 2.78mm) by means of further processing. The bone contour was found with high accuracy. Median error orthogonal to the contour was 0.113mm with T30. With T30 74.6% of the landmarks in the test set lie within a 0.25mm error corridor around the true bone contour. Overlapping bones slightly deteriorate the result. Features extracted from the algorithm show good discriminative properties for the automated detection of bone erosions.

Conclusion: A feasible and robust way to initialise automated quantification of bone lesions and a method for the automated estimation of the bone contour is proposed. The result of this algorithm is a parametrical description of the identified shape. This method is not only applicable for hand radiographs but also for image analysis in other radiological tasks. Mathematically extracted features leading to optimal results in the detection of bone erosions are to be evaluated in ongoing studies. The research has been partly supported by Austrian Science Fund (FWF) under grant P14445-MAT and P14662-INF.




Questions about this event email: